In the United Kingdom, a national habit can unexpectedly threaten the stability of the electrical grid. During the half-time whistle of the UEFA EURO 2020 round of 16 match between England and Germany, millions of viewers simultaneously stepped away from their screens to boil their kettles. This collective action triggered a demand spike of approximately 1 gigawatt—an increase equivalent to the average output of a standard nuclear reactor—within a matter of minutes.
For grid operators, these “TV pickup” events are high-stress scenarios that require precise management to prevent system failure. Yet, as the global race for artificial intelligence accelerates, the grid faces a new challenge: the arrival of “AI factories”—massive data centers that consume vast amounts of power. Traditionally, these facilities have been viewed as burdens on the energy infrastructure, often facing years-long delays for grid connections although waiting for expensive hardware upgrades.
A new approach to power-flexible AI factories suggests that these facilities could transition from being power drains to becoming active stabilizers for the global energy grid. By autonomously adjusting their energy consumption in real-time, AI factories can act as a “shock absorber” for the grid, relieving strain during peak demand and potentially lowering electricity rates for the general public by reducing the need for permanent, overbuilt infrastructure.
This shift in utility was detailed in a recent white paper produced by Emerald AI in collaboration with NVIDIA, the Electric Power Research Institute (EPRI), National Grid, and Nebius. The research demonstrates a system where AI workloads are not just consumers of energy, but flexible assets that can be throttled up or down based on the immediate needs of the power grid.
The London Experiment: Simulating the National Tea Break
Following successful proof-of-concept trials in Arizona, Virginia, and Illinois, Emerald AI deployed its Conductor Platform last December at a new Nebius AI factory in London. This facility, built on NVIDIA infrastructure, served as a testing ground to see if an AI cluster could react to grid stress signals without crashing critical operations.
The research team utilized a cluster of 96 NVIDIA Blackwell Ultra GPUs, connected via the NVIDIA Quantum-X800 InfiniBand platform. To ensure precision, the team used the NVIDIA System Management Interface to retrieve GPU power telemetry at second-level intervals. This allowed the Conductor Platform to make near-instantaneous adjustments to power draw.
EPRI and National Grid simulated a variety of grid stress events, ranging from lightning strikes to periods of low wind power supply. The most notable test was the reenactment of the Euro 2020 “TV pickup” phenomenon. As the simulated demand for tea kettles surged, the AI cluster ramped down its power usage, successfully absorbing the shock of the surge without disrupting high-priority AI workloads.
Balancing Throughput and Grid Stability
The primary technical hurdle for power-flexible AI factories is ensuring that reducing energy does not result in “query crushing”—the failure of essential AI tasks. The Nebius demonstration addressed this by categorizing workloads. High-priority tasks maintained peak throughput, while more flexible, non-urgent jobs were temporarily slowed down.
The results indicated a high level of precision in energy management. Emerald AI reported 100% alignment with over 200 specific power targets set by EPRI and National Grid. This level of control extends beyond the GPUs; the testing included CPUs and the surrounding IT equipment to ensure total power consumption was managed.
“We did tests that go beyond the ones that have been done so far in the U.S. Because we tested not just the GPUs, but also the CPUs and everything that sits around it — as well as the total power consumption of the IT equipment,” said Steve Smith, group chief strategy officer of National Grid. “We’ve proved the value that this technology brings.”
Impact on Infrastructure and Economics
The implications of this technology extend beyond technical stability to the economic viability of AI deployment. In many urban centers, including London, the primary bottleneck for new industry is the constraint of existing grid infrastructure. Upgrading these systems can take years and cost billions.
By utilizing power-flexible AI factories, companies can potentially bypass these lengthy upgrade cycles. If a facility can prove it will not exacerbate peak loads, grid operators may be able to grant faster connections using existing capacity.
| Stakeholder | Primary Benefit | Long-term Impact |
|---|---|---|
| AI Factory Operators | Faster grid connection | Reduced time-to-market for AI services |
| Grid Operators | Peak load mitigation | Reduced need for emergency backup power |
| General Public | Stabilized energy rates | Lower costs by avoiding massive infrastructure overbuilds |
| National Economy | Industrial growth | Increased capacity for hyperscalers and AI talent |
Unlocking Economic Growth in the U.K.
For the United Kingdom, this technology represents a strategic opportunity to compete with the United States in the AI sector. While the U.K. May not match the absolute scale of U.S. Data center clusters, the ability to integrate AI factories efficiently into the existing grid allows for a higher density of innovation relative to the country’s size.
Steve Smith of National Grid noted that there is significant interest from hyperscalers to expand in the region. By transforming AI factories into “friendly and helpful grid assets,” as described by Emerald AI founder and CEO Varun Sivaram, the U.K. Can leverage its existing AI skills and potential without crashing its power systems.
The transition from theoretical white papers to real-world application is already underway. Following the four successful demonstrations, Emerald AI and NVIDIA are moving toward a full-scale deployment at the Aurora AI Factory in Virginia, which is scheduled to open this year.
As AI continues to scale, the industry’s success will depend not only on the number of GPUs available but on the ability to integrate those chips into a fragile global energy ecosystem. The move toward flexibility marks a shift from AI as a consumer of resources to AI as a partner in infrastructure management.
We invite readers to share their thoughts on the intersection of AI and energy sustainability in the comments below.
